spike - a command line tool for continuous, stochastic ... · [email protected] awpn october...

[email protected] AWPN October 2018 01

Spike - a command line tool forcontinuous, stochastic & hybrid simulation

of (coloured) Petri nets

Jacek Chodak, Monika Heiner (supervisor)Computer Science Institute, Brandenburg University of Technology

Cottbus, [email protected], [email protected]

http://www-dssz.informatik.tu-cottbus.de/DSSZ/Software/Spike


Agenda

* PetriNuts framework* Why Spike* Spike functionality - Simulation - Conversion - Reduction - Reproducibility* Architecture* Use cases* Present status* Future works


PetriNuts Framework

CHARLIE [1] • Structural analysis • Analyses building on incidence matrix • Reachability graph analysis • Siphon/trap computation • Explicite CTL model checking • Structural reduction

SNOOPY [3, 4]• Modelling• Simulation: - stochastic, - continuous, - hybrid• Animation

MARCIE [2]• Symbolic analysis (PN) - state space analysis - CTL model checking• Symbolic analysis (SPN) - standard numerical analysis, - CSRL model checking• Simulative analysis (SPN) - standard numerical analysis - PLTc model checking • explicit approximative numerical analysis

SPIKE• Distributed simulation: - stochastic, - continuous, - hybrid

PATTY [5]• Web based animation

References:• [1] M Heiner, M Schwarick and J Wegener: PETRI NETS 2015• [2] M Heiner, C Rohr and M Schwarick: PETRI NETS 2013• [3] M Herajy, F Liu, C Rohr and M Heiner: BMC Systems Biology 2017• [4] C Rohr, W Marvan, M Heiner: Bioinformatics 2010• [5] K Schulz, BA thesis, BTU Cottbus, CS department 2008


Why Spike

* Reproducibility

* Simplifying workflow of simulation experiments

* Speedup simulation Models can contain dozens of thousands of nodes. To speed up simulation, a model can be

reduced or divided into modules (spatial decomposition, decomposition by node types) and simulated in a distributed way.

There are many parameters: model parameters (e.g., initial marking, kinetic constants); simulations

parameters(e.g., type of algm, length of trace, number of stochastic runs). Experiments with Spike are documented by configuration files.

CLI controlling multiple simulations without user intervention; typically for different model configurations and/or simulator configurations.


Simulation

* CPN - Continuous Petri Net (ODEs)* HPN - Hybrid Petri Net

SPN CPN

HPN

<=>

<=>

<=>

Supported (coloured) PN classes:* SPN - Stochastic Petri Net (CONTINUOUS TIME MARKOV CHAINS)


Simulation

* SPN - Stochastic Petri Net (CONTINUOUS TIME MARKOV CHAINS)* CPN - Continuous Petri Net (ODEs)* HPN - Hybrid Petri Net

SPN CPN

HPN

<=>

<=>

<=>

Supported (coloured) PN classes:


Simulation

* Stochastic

* Continuous

* Hybrid

time

time

time

CPN, HPN => SPN

SPN, HPN => CPN

SPN, CPN => HPN


Conversion

ANDL

CANDL

SBML

PNML

ERODE

human readable formats for Petri nets and Coloured Petri nets, respectively, used internally by the PetriNuts framework

(the Systems Biology Markup Language) - an XML-based representation format designed to exchange computational models of biological processes

a tool for the evaluation and reduction of chemical reactionnetworks readed as ordinary differential equations (ODEs)

an XML-based interchange format for Petri nets (no time)


Conversion

ANDL

CANDLSBML

PNML

ERODE

SPIKE ANDL

ERODE

Import Export


ReductionModel

Reduced Model

Reduction* Clean siphons* Constant places* ERODE Further reductions may be

applied by converting a model to the ERODE format, if the model is to be read as ordinary differential equations (ODEs).

* further general reduction rules under consideration


Reproducibilityconfiguration: { model: { constants: { ... } places: { ... } observers: { ... } }

* model configuration** constants

** observers** initial states(arc weights, initial markings, kinetic constants)

* simulation configurations** exports

simulation : { ... export: { ... } } ...}


Reproducibility

* Definition of constantsconstants: { // name of a group all: { /∗ if constant does not exist ∗ then it will be created and ∗ can be used in the configuration , ∗ for example in defining a place marking ∗/ M: ”D/2 + 1” }}


Reproducibility

places: {// example of use of the newly created constant MP: "1000`(M,M)"P_2_2: 500

}

* Set marking for places


Reproducibility

observers: {place: {OP01: {function: "P_1_1 + P_2_3"

}}

}

* Definition of auxiliary variables (observers) which allow for extra measures by defining numerical functions; depending on the type of observer, it can be defined for places, transitions or simultaneously for places, transitions and constants


Reproducibility

simulation: { name: "Diffusion" // Name of a simulation type: stochastic // [continuous, stochastic, hybrid] solver: direct // [BDF, classic, direct, delta-leaping, static, staticAcc] threads: 0 // 0 - try to get the number of concurrent threads supported by the hardware interval: 1:1:1000// start : step : end (startTimePoint : intervalSize : endTimePoint) runs: 10 // Applicable only for stochastic sim // continues conf iniStep: 0.1 linSolver: CVDense //[CVDense, CVSpgmr, CVDiag, CVSpbcg, and CVSptfqmr] relTol: 1.0e-5 absTol: 1.0e-10 // Export export: { places: [] // Array of places to save (if empty export all) transitions: [] to: "testCase-diffusion-stochastic.csv" }}

* Defining multiple simulation configurations, which permits to run multiple experiments for one model configuration

//simulation: { ...}


Reproducibility

export: {// Array of places to save,// including colored places like P// in this example (if empty, export all)places: ["P_1_1", "OP01", "Grid.*", "D", "P"]// Array of transitions to save,// including colored transitionstransitions: ["t3_1_1_1_2", "t3", "t3" ]// Array of observers to save (if empty, export all)observers: ["M01", "OT01"]to: "sim01-file01.csv"

}export: {...}

* Defining multiple exports of simulation results by use of regular expressions over the nodes; any combination of places, transitions and observers, coloured and uncoloured, in one file


Architecture

* Command queue, a typical use case


Architecture


Command queue


Architecture


CLICommand queueexe

user cmd: spike exe -f confFile


Architecture


CLI exeCommand queueexe



Architecture


CLI exeCommand queue

Configurationexe exe-



Architecture


CLI exeload

Command queueConfiguration

exe exeload-



Architecture


CLI exeload

unfold


exe exeunfold-



Architecture


CLI exeload

unfoldeval


exe exeeval-



Architecture


CLI exeload

unfoldevalsim


exe exesim-



Architecture


CLI exeload

unfoldevalsim

Command queueConfigurationload, unfold, eval, sim

Converter

exe exe

load

--



Architecture


CLI exeload

unfoldevalsim


Converter

exe exe

unfold---



Architecture


CLI exeload

unfoldevalsim


Converter

exe exe

eval----



Architecture


CLI exeload

unfoldevalsim


ConverterSimulation

load, unfold, eval

exe exe

sim

-----



Use case

1 Load model2 Set model configuration3 Determine set of simulation configurations4 For( each simulation configuration ) {5 Run simulation6 Save results of the simulation7 }

Compare how a model behaves under different types of simulation algorithms or under different configurations of a given simulation algorithm.


Present status

Mode

l a

nd simulation configuration

Model

Reduced Model

Deco

mpos

ition

Reduction

Components

Finding

interface

nodes

Distribution

Distributed model

synchronization

Simulation

Co

mponen

ts with interfaces

Traces

Reproduciblesimulation

Export

Redu

ction

TraceAnalysis


Future works

Mode

l a


Model

Reduced Model

Deco

mpos

ition

Reduction

Components

Finding

interface

nodes

Distribution

Distributed model

synchronization

Simulation

Co

mponen

ts with interfaces

Traces


Export

Redu

ction

TraceAnalysis

X


Future works

Mode

l a


Model

Reduced Model

Deco

mpos

ition

Reduction

Components

Finding

interface

nodes

Distribution

Distributed model

synchronization

Simulation

Co

mponen

ts with interfaces

Traces


Export

Redu

ction

TraceAnalysis

X

X

X


Future works

Mode

l a


Model

Reduced Model

Deco

mpos

ition

Reduction

Components

Finding

interface

nodes

Distribution

Distributed model

synchronization

Simulation

Co

mponen

ts with interfaces

Traces


Export

Redu

ction

TraceAnalysis

X

X

XX


Thank you for your attention

http://www-dssz.informatik.tu-cottbus.de/DSSZ/Software/Spike

Any questions are welcome

spike - a command line tool for continuous, stochastic ... · [email protected] awpn october...

Documents